Blob features are usually used as texturedescriptors for object classification and most traditionalblob detectors nowadays are luminance-based. However,in many applications we could require a blob detector forthe color domain. In this paper, we propose a novel andefficient framework for the extension from scalar-signals tovector-signals of multi-scale blob detection to prevent informationloss due to gray scale transformation. Then, wepresent a blob detector for stamp image classification todistinguish the stamp types. In our experiments, we compareour method to other approaches. The experimentalresults demonstrate the effectiveness of our proposed detector.